def svm_cv(argv):
"""A top level script to parse input parameters and run cross validation"""
assert(argv[1]=='cv')
if len(argv)<5:sys.stderr.write("usage: %s cv repeat C kernelname [kernelparameters] [arff|fasta] inputfiles outputfile [dna|protein] non(nucleotide|amino)converter \n" % argv[0]);sys.exit(-1)
# parse input parameters
cv = int(argv[2])
C = float(argv[3])
(kernelname,kparam,argv_rest) = parse.parse_kernel_param(argv[4:],False)
(examples,labels,argv_rest) = parse.parse_input_file_train(kernelname, argv_rest)
(seq_source, nuc_con) = ('', '')
if kernelname == 'spec' or kernelname == 'wd':
if len(argv_rest)<1:sys.stderr.write("outputfile [dna|protein] non(nucleotide|amino)converter are missing\n");sys.exit(-1)
if len(argv_rest)<2:sys.stderr.write("[dna|protein] non(nucleotide|amino)converter are missing\n");sys.exit(-1)
if len(argv_rest)<3:
if argv_rest[-1] == 'dna':
sys.stderr.write("non-nucleotide converter like [A|T|C|G|R|Y|N] is missing. Cannot continue.\n")
sys.exit(-1)
elif argv_rest[-1] == 'protein':
sys.stderr.write("non-amino acid converter like [G|P|A|V|L|I|M|C|F|Y|W|H|K|R|Q|N|E|D|S|T|random] is missing. Cannot continue.\n")
sys.exit(-1)
else:
sys.stderr.write("Here expect FASTA sequence type as [dna|protein] instead of -"+ argv_rest[-1] +"- Cannot continue.\n")
sys.exit(-1)
if len(argv_rest)>3:sys.stderr.write("Too many arguments\n");sys.exit(-1)
seq_source = argv_rest[1]
nuc_con = argv_rest[2]
if kernelname == 'linear' or kernelname == 'gauss' or kernelname == 'poly':
if len(argv_rest)<1:sys.stderr.write("outputfile misssing\n");sys.exit(-1)
if len(argv_rest)>1:sys.stderr.write("Too many arguments\n");sys.exit(-1)
outfilename = argv_rest[0]
utils.check_params(kparam, C, len(examples[0]))
# run cross-validation
(all_outputs, all_split) = crossvalidation(cv, kernelname, kparam, C, examples, labels, seq_source, nuc_con)
try:
f = open(outfilename, 'w+')
except:
sys.stderr.write('Fails to open the outputfile at ' + outfilename + ' Cannot continue.\n')
sys.exit(-1)
res_str = '#example\toutput\tsplit\n'
f.write(res_str)
for ix in xrange(len(all_outputs)):
res_str = '%d\t%2.7f\t%d\n' % (ix,all_outputs[ix],all_split[ix])
f.write(res_str)
f.close()
def svm_eval(argv):
"""A top level script to parse input parameters and evaluate"""
assert (argv[1] == 'eval')
if len(argv) < 6:
sys.stderr.write(
"usage: %s eval predictionfile [arff|fasta] inputfiles outputfile [roc|prc figure.png]\n"
% argv[0])
sys.exit(-1)
# parse input parameters
(predictions, splitassignments) = parse.parse_prediction_file(argv[2])
(trainex, trainlab,
argv_rest) = parse.parse_input_file_train(None, argv[3:])
if len(argv_rest) < 1:
sys.stderr.write("Output file missing\n")
sys.exit(-1)
if len(argv_rest) > 3:
sys.stderr.write("Too many arguments\n")
sys.exit(-1)
outfilename = argv_rest[0]
roc_fname = None
prc_fname = None
if len(argv_rest) > 2:
if argv_rest[1] == 'roc':
roc_fname = argv_rest[2]
elif argv_rest[1] == 'prc':
prc_fname = argv_rest[2]
else:
sys.stderr.write('Usage: [roc|prc]')
sys.exit(-1)
# run training and testing
(res_str, mean_roc, mean_prc,
mean_acc) = evaluate(predictions, splitassignments, trainlab, roc_fname,
prc_fname)
# write output file
try:
f = open(outfilename, 'w')
except:
sys.stderr.write('Fails to open the outputfile at ' + outfilename +
' Cannot continue.\n')
sys.exit(-1)
f.write(res_str)
f.close()
def svm_poim(argv):
"""A top level script to parse input parameters and plot poims"""
assert(argv[1]=='poim')
if len(argv)<7:sys.stderr.write("usage: %s poim C poimdegree wd [kernelparameters] [arff|fasta] inputfiles poim.png [dna|protein] non(nucleotide|amino)converter\n" % argv[0]);sys.exit(-1)
# parse input parameters
C = float(argv[2])
poimdegree = int(argv[3])
(kernelname,kparam,argv_rest) = parse.parse_kernel_param(argv[4:], False)
(examples,labels,argv_rest) = parse.parse_input_file_train(kernelname, argv_rest)
if len(argv_rest)<1:sys.stderr.write("poim.png [dna|protein] non(nucleotide|amino)converter are missing\n");sys.exit(-1)
if len(argv_rest)<2:sys.stderr.write("[dna|protein] non(nucleotide|amino)converter are missing\n");sys.exit(-1)
if len(argv_rest)<3:
if argv_rest[-1] == 'dna':
sys.stderr.write("non-nucleotide converter like [A|T|C|G|R|Y|N] is missing. Cannot continue.\n")
sys.exit(-1)
elif argv_rest[-1] == 'protein':
sys.stderr.write("non-amino acid converter like [G|P|A|V|L|I|M|C|F|Y|W|H|K|R|Q|N|E|D|S|T|random] is missing. Cannot continue.\n")
sys.exit(-1)
else:
sys.stderr.write("Here expect FASTA sequence type as [dna|protein] instead of -"+ argv_rest[-1] +"- Cannot continue.\n")
sys.exit(-1)
if len(argv_rest)>3:sys.stderr.write("Too many arguments\n");sys.exit(-1)
poimfilename = argv_rest[0]
seq_source = argv_rest[1]
nuc_con = argv_rest[2]
utils.check_params(kparam, C, len(examples[0]))
# train svm and compute POIMs
(svm, kernel, feats_train, preproc) = train(examples,labels,C,kernelname,kparam,seq_source,nuc_con)
print "done with training "
(poim, max_poim, diff_poim, poim_totalmass) = compute_poims(svm, kernel, poimdegree, len(examples[0]))
# plot poims
plots.plot_poims(poimfilename, poim, max_poim, diff_poim, poim_totalmass, poimdegree, len(examples[0]))
def svm_poim(argv):
"""A top level script to parse input parameters and plot poims"""
assert(argv[1]=='poim')
if len(argv)<7:sys.stderr.write("usage: %s poim C poimdegree wd [kernelparameters] [arff|fasta] inputfiles poim.png [dna|protein] non(nucleotide|amino)converter\n" % argv[0]);sys.exit(-1)
# parse input parameters
C = float(argv[2])
poimdegree = int(argv[3])
(kernelname,kparam,argv_rest) = parse.parse_kernel_param(argv[4:], False)
(examples,labels,argv_rest) = parse.parse_input_file_train(kernelname, argv_rest)
if len(argv_rest)<1:sys.stderr.write("poim.png [dna|protein] non(nucleotide|amino)converter are missing\n");sys.exit(-1)
if len(argv_rest)<2:sys.stderr.write("[dna|protein] non(nucleotide|amino)converter are missing\n");sys.exit(-1)
if len(argv_rest)<3:
if argv_rest[-1] == 'dna':
sys.stderr.write("non-nucleotide converter like [A|T|C|G|R|Y|N] is missing. Cannot continue.\n")
sys.exit(-1)
elif argv_rest[-1] == 'protein':
sys.stderr.write("non-amino acid converter like [G|P|A|V|L|I|M|C|F|Y|W|H|K|R|Q|N|E|D|S|T|random] is missing. Cannot continue.\n")
sys.exit(-1)
else:
sys.stderr.write("Here expect FASTA sequence type as [dna|protein] instead of -"+ argv_rest[-1] +"- Cannot continue.\n")
sys.exit(-1)
if len(argv_rest)>3:sys.stderr.write("Too many arguments\n");sys.exit(-1)
poimfilename = argv_rest[0]
seq_source = argv_rest[1]
nuc_con = argv_rest[2]
utils.check_params(kparam, C, len(examples[0]))
# train svm and compute POIMs
(svm, kernel, feats_train, preproc) = train(examples,labels,C,kernelname,kparam,seq_source,nuc_con)
(poim, max_poim, diff_poim, poim_totalmass) = compute_poims(svm, kernel, poimdegree, len(examples[0]))
# plot poims
plots.plot_poims(poimfilename, poim, max_poim, diff_poim, poim_totalmass, poimdegree, len(examples[0]))
def svm_eval(argv):
"""A top level script to parse input parameters and evaluate"""
assert(argv[1]=='eval')
if len(argv)<6:sys.stderr.write("usage: %s eval predictionfile [arff|fasta] inputfiles outputfile [roc|prc figure.png]\n" % argv[0]);sys.exit(-1)
# parse input parameters
(predictions, splitassignments) = parse.parse_prediction_file(argv[2])
(trainex, trainlab, argv_rest) = parse.parse_input_file_train(None, argv[3:])
if len(argv_rest)<1:sys.stderr.write("Output file missing\n");sys.exit(-1)
if len(argv_rest)>3:sys.stderr.write("Too many arguments\n");sys.exit(-1)
outfilename = argv_rest[0]
roc_fname = None
prc_fname = None
if len(argv_rest)>2:
if argv_rest[1]=='roc':
roc_fname=argv_rest[2]
elif argv_rest[1]=='prc':
prc_fname=argv_rest[2]
else:
sys.stderr.write('Usage: [roc|prc]')
sys.exit(-1)
# run training and testing
(res_str,mean_roc,mean_prc,mean_acc) = evaluate(predictions, splitassignments, trainlab, roc_fname, prc_fname)
# write output file
try:
f = open(outfilename,'w')
except:
sys.stderr.write('Fails to open the outputfile at ' + outfilename + ' Cannot continue.\n')
sys.exit(-1)
f.write(res_str)
f.close()
def svm_modelsel(argv):
"""A top level script to parse input parameters and run model selection"""
assert(argv[1]=='modelsel')
if len(argv)<5:sys.stderr.write("usage: %s modelsel repeat Cs kernelname [kernelparameters] [arff|fasta] inputfiles outputfile [dna|protein] non(nucleotide|amino)converter\n" % argv[0]);sys.exit(-1)
# parse input parameters
cv = int(argv[2])
Cs = parse.parse_float_list(argv[3])
(kernelname,kparam,argv_rest) = parse.parse_kernel_param(argv[4:], True)
(examples,labels,argv_rest) = parse.parse_input_file_train(kernelname, argv_rest)
(seq_source, nuc_con) = ('', '')
if kernelname == 'spec' or kernelname == 'wd':
if len(argv_rest)<1:sys.stderr.write("outputfile [dna|protein] non(nucleotide|amino)converter are missing\n");sys.exit(-1)
if len(argv_rest)<2:sys.stderr.write("[dna|protein] non(nucleotide|amino)converter are missing\n");sys.exit(-1)
if len(argv_rest)<3:
if argv_rest[-1] == 'dna':
sys.stderr.write("non-nucleotide converter like [A|T|C|G|R|Y|N] is missing. Cannot continue.\n")
sys.exit(-1)
elif argv_rest[-1] == 'protein':
sys.stderr.write("non-amino acid converter like [G|P|A|V|L|I|M|C|F|Y|W|H|K|R|Q|N|E|D|S|T|random] is missing. Cannot continue.\n")
sys.exit(-1)
else:
sys.stderr.write("Here expect FASTA sequence type as [dna|protein] instead of -"+ argv_rest[-1] +"- Cannot continue.\n")
sys.exit(-1)
if len(argv_rest)>3:sys.stderr.write("Too many arguments\n");sys.exit(-1)
seq_source = argv_rest[1]
nuc_con = argv_rest[2]
if kernelname == 'linear' or kernelname == 'gauss' or kernelname== 'poly':
if len(argv_rest)<1:sys.stderr.write("outputfile missing\n");sys.exit(-1)
if len(argv_rest)>1:sys.stderr.write("Too many arguments\n");sys.exit(-1)
outfilename = argv_rest[0]
# run cross-validation
mean_rocs=[] ;
mean_prcs=[] ;
mean_accs=[] ;
all_Cs = [] ;
all_kparam=[] ;
if kparam["modelsel_name"]==None:
for C in Cs:
utils.check_params(kparam, C, len(examples[0]))
(all_outputs, all_split) = crossvalidation(cv, kernelname, kparam, C, examples, labels, seq_source, nuc_con)
(res_str, mean_roc, mean_prc, mean_acc) = evaluate(all_outputs, all_split, labels)
mean_rocs.append(mean_roc)
mean_prcs.append(mean_prc)
mean_accs.append(mean_acc)
all_Cs.append(C)
all_kparam.append(None)
else: # also optimize one kernel parameter
for C in Cs:
for kp in kparam["modelsel_params"]:
kparam[kparam["modelsel_name"]] = kp
utils.check_params(kparam, C, len(examples[0]))
(all_outputs, all_split) = crossvalidation(cv, kernelname, kparam, C, examples, labels, seq_source, nuc_con)
(res_str, mean_roc, mean_prc, mean_acc) = evaluate(all_outputs, all_split, labels)
mean_rocs.append(mean_roc)
mean_prcs.append(mean_prc)
mean_accs.append(mean_acc)
all_Cs.append(C)
all_kparam.append(kp)
max_roc=numpy.max(numpy.array(mean_rocs))
max_prc=numpy.max(numpy.array(mean_prcs))
max_acc=numpy.max(numpy.array(mean_accs))
try:
f = open(outfilename, 'w+')
except:
sys.stderr.write('Fails to open the outputfile at ' + outfilename + ' Cannot continue.\n')
sys.exit(-1)
if kparam["modelsel_name"]==None or len(kparam["modelsel_params"])==1:
detail_str = "\tC\tROC\tPRC\tAccuracy (at threshold 0)\n"
else:
detail_str = "\tC\t%s\tROC\tPRC\tAccuracy (at threshold 0)\n" % kparam["modelsel_name"]
best_roc_str=''
best_prc_str=''
best_acc_str=''
for i in xrange(len(all_Cs)):
# determine the best parameter combinations
if mean_rocs[i]==max_roc:
rocsym='+'
best_roc_str+=model2str(kparam, all_Cs[i], all_kparam[i])+'\n'
else:
rocsym=' '
if mean_prcs[i]==max_prc:
prcsym='+'
best_prc_str+=model2str(kparam, all_Cs[i], all_kparam[i])+'\n'
else:
prcsym=' '
if mean_accs[i]==max_acc:
accsym='+'
best_acc_str+=model2str(kparam, all_Cs[i], all_kparam[i])+'\n'
else:
accsym=' '
detail_str+=model2str(kparam, all_Cs[i], all_kparam[i], False)+'\t'
if kparam["modelsel_name"]==None or len(kparam["modelsel_params"])==1:
detail_str += '%c%2.1f%%\t%c%2.1f%%\t%c%2.1f%%\n' % (rocsym, 100*mean_rocs[i], prcsym, 100*mean_prcs[i], accsym, 100*mean_accs[i])
else:
detail_str += '%c%2.1f%%\t%c%2.1f%%\t%c%2.1f%%\n' % (rocsym, 100*mean_rocs[i], prcsym, 100*mean_prcs[i], accsym, 100*mean_accs[i])
f.write('Best model(s) according to ROC measure:\n%s' % best_roc_str)
f.write('\nBest model(s) according to PRC measure:\n%s' % best_prc_str)
f.write('\nBest model(s) according to accuracy measure:\n%s' % best_acc_str)
f.write('\nDetailed results:\n')
f.write(detail_str)
f.close()